4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
36 #include <sys/zfs_context.h>
37 #include <sys/zfs_refcount.h>
38 #include <sys/zrlock.h>
39 #include <sys/multilist.h>
48 * define flags for dbuf_read
51 #define DB_RF_MUST_SUCCEED (1 << 0)
52 #define DB_RF_CANFAIL (1 << 1)
53 #define DB_RF_HAVESTRUCT (1 << 2)
54 #define DB_RF_NOPREFETCH (1 << 3)
55 #define DB_RF_NEVERWAIT (1 << 4)
56 #define DB_RF_CACHED (1 << 5)
57 #define DB_RF_NO_DECRYPT (1 << 6)
60 * The simplified state transition diagram for dbufs looks like:
65 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
71 * +--------> NOFILL -------+
73 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
74 * to find all dbufs in a range of a dnode and must be less than any other
75 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
77 typedef enum dbuf_states {
87 typedef enum dbuf_cached_state {
90 DB_DBUF_METADATA_CACHE,
92 } dbuf_cached_state_t;
98 * level = 0 means the user data
99 * level = 1 means the single indirect block
105 typedef enum override_states {
111 typedef enum db_lock_type {
117 typedef struct dbuf_dirty_record {
118 /* link on our parents dirty list */
119 list_node_t dr_dirty_node;
121 /* transaction group this data will sync in */
124 /* zio of outstanding write IO */
127 /* pointer back to our dbuf */
128 struct dmu_buf_impl *dr_dbuf;
130 /* list link for dbuf dirty records */
131 list_node_t dr_dbuf_node;
134 * The dnode we are part of. Note that the dnode can not be moved or
135 * evicted due to the hold that's added by dnode_setdirty() or
136 * dmu_objset_sync_dnodes(), and released by dnode_rele_task() or
137 * userquota_updates_task(). This hold is necessary for
138 * dirty_lightweight_leaf-type dirty records, which don't have a hold
143 /* pointer to parent dirty record */
144 struct dbuf_dirty_record *dr_parent;
146 /* How much space was changed to dsl_pool_dirty_space() for this? */
147 unsigned int dr_accounted;
149 /* A copy of the bp that points to us */
153 struct dirty_indirect {
155 /* protect access to list */
158 /* Our list of dirty children */
164 * dr_data is set when we dirty the buffer
165 * so that we can retain the pointer even if it
166 * gets COW'd in a subsequent transaction group.
169 blkptr_t dr_overridden_by;
170 override_states_t dr_override_state;
172 boolean_t dr_nopwrite;
173 boolean_t dr_has_raw_params;
176 * If dr_has_raw_params is set, the following crypt
177 * params will be set on the BP that's written.
179 boolean_t dr_byteorder;
180 uint8_t dr_salt[ZIO_DATA_SALT_LEN];
181 uint8_t dr_iv[ZIO_DATA_IV_LEN];
182 uint8_t dr_mac[ZIO_DATA_MAC_LEN];
184 struct dirty_lightweight_leaf {
186 * This dirty record refers to a leaf (level=0)
187 * block, whose dbuf has not been instantiated for
188 * performance reasons.
193 enum zio_flag dr_flags;
196 } dbuf_dirty_record_t;
198 typedef struct dmu_buf_impl {
200 * The following members are immutable, with the exception of
201 * db.db_data, which is protected by db_mtx.
204 /* the publicly visible structure */
207 /* the objset we belong to */
208 struct objset *db_objset;
211 * handle to safely access the dnode we belong to (NULL when evicted)
213 struct dnode_handle *db_dnode_handle;
216 * our parent buffer; if the dnode points to us directly,
217 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
218 * only accessed by sync thread ???
219 * (NULL when evicted)
220 * May change from NULL to non-NULL under the protection of db_mtx
221 * (see dbuf_check_blkptr())
223 struct dmu_buf_impl *db_parent;
226 * link for hash table of all dmu_buf_impl_t's
228 struct dmu_buf_impl *db_hash_next;
231 * Our link on the owner dnodes's dn_dbufs list.
232 * Protected by its dn_dbufs_mtx. Should be on the same cache line
233 * as db_level and db_blkid for the best avl_add() performance.
237 /* our block number */
241 * Pointer to the blkptr_t which points to us. May be NULL if we
242 * don't have one yet. (NULL when evicted)
247 * Our indirection level. Data buffers have db_level==0.
248 * Indirect buffers which point to data buffers have
249 * db_level==1. etc. Buffers which contain dnodes have
250 * db_level==0, since the dnodes are stored in a file.
255 * Protects db_buf's contents if they contain an indirect block or data
256 * block of the meta-dnode. We use this lock to protect the structure of
257 * the block tree. This means that when modifying this dbuf's data, we
258 * grab its rwlock. When modifying its parent's data (including the
259 * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering
261 * 1) dn_struct_rwlock
263 * We don't currently grab multiple dbufs' db_rwlocks at once.
267 /* buffer holding our data */
270 /* db_mtx protects the members below */
274 * Current state of the buffer
276 dbuf_states_t db_state;
279 * Refcount accessed by dmu_buf_{hold,rele}.
280 * If nonzero, the buffer can't be destroyed.
281 * Protected by db_mtx.
283 zfs_refcount_t db_holds;
285 kcondvar_t db_changed;
286 dbuf_dirty_record_t *db_data_pending;
288 /* List of dirty records for the buffer sorted newest to oldest. */
289 list_t db_dirty_records;
291 /* Link in dbuf_cache or dbuf_metadata_cache */
292 multilist_node_t db_cache_link;
294 /* Tells us which dbuf cache this dbuf is in, if any */
295 dbuf_cached_state_t db_caching_status;
297 /* Data which is unique to data (leaf) blocks: */
299 /* User callback information. */
300 dmu_buf_user_t *db_user;
303 * Evict user data as soon as the dirty and reference
306 uint8_t db_user_immediate_evict;
309 * This block was freed while a read or write was
312 uint8_t db_freed_in_flight;
315 * dnode_evict_dbufs() or dnode_evict_bonus() tried to
316 * evict this dbuf, but couldn't due to outstanding
317 * references. Evict once the refcount drops to 0.
319 uint8_t db_pending_evict;
324 /* Note: the dbuf hash table is exposed only for the mdb module */
325 #define DBUF_MUTEXES 2048
326 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
327 typedef struct dbuf_hash_table {
328 uint64_t hash_table_mask;
329 dmu_buf_impl_t **hash_table;
330 kmutex_t hash_mutexes[DBUF_MUTEXES] ____cacheline_aligned;
333 typedef void (*dbuf_prefetch_fn)(void *, uint64_t, uint64_t, boolean_t);
335 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level,
336 const uint64_t offset);
338 void dbuf_create_bonus(struct dnode *dn);
339 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
341 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
343 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, const void *tag);
344 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
346 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
347 boolean_t fail_sparse, boolean_t fail_uncached,
348 const void *tag, dmu_buf_impl_t **dbp);
350 int dbuf_prefetch_impl(struct dnode *dn, int64_t level, uint64_t blkid,
351 zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb,
353 int dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
354 zio_priority_t prio, arc_flags_t aflags);
356 void dbuf_add_ref(dmu_buf_impl_t *db, const void *tag);
357 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
358 uint64_t blkid, const void *tag);
359 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
361 void dbuf_rele(dmu_buf_impl_t *db, const void *tag);
362 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, const void *tag,
365 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
368 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
369 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
370 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
371 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
372 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
373 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
374 dbuf_dirty_record_t *dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid,
376 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
377 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
378 bp_embedded_type_t etype, enum zio_compress comp,
379 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
381 int dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd,
382 const struct zio_prop *zp, enum zio_flag flags, dmu_tx_t *tx);
384 void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx);
385 void dbuf_destroy(dmu_buf_impl_t *db);
387 void dbuf_unoverride(dbuf_dirty_record_t *dr);
388 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
389 void dbuf_release_bp(dmu_buf_impl_t *db);
390 db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw,
392 void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type,
395 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
398 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
400 void dbuf_stats_init(dbuf_hash_table_t *hash);
401 void dbuf_stats_destroy(void);
403 int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid,
404 blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift);
406 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
407 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
408 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
409 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
410 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
412 void dbuf_init(void);
413 void dbuf_fini(void);
415 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
417 static inline dbuf_dirty_record_t *
418 dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg)
420 dbuf_dirty_record_t *dr;
422 for (dr = list_head(&db->db_dirty_records);
423 dr != NULL && dr->dr_txg > txg;
424 dr = list_next(&db->db_dirty_records, dr))
429 static inline dbuf_dirty_record_t *
430 dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg)
432 dbuf_dirty_record_t *dr;
434 dr = dbuf_find_dirty_lte(db, txg);
435 if (dr && dr->dr_txg == txg)
440 #define DBUF_GET_BUFC_TYPE(_db) \
441 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
443 #define DBUF_IS_CACHEABLE(_db) \
444 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
445 (dbuf_is_metadata(_db) && \
446 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
448 boolean_t dbuf_is_l2cacheable(dmu_buf_impl_t *db);
453 * There should be a ## between the string literal and fmt, to make it
454 * clear that we're joining two strings together, but gcc does not
455 * support that preprocessor token.
457 #define dprintf_dbuf(dbuf, fmt, ...) do { \
458 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
460 uint64_t __db_obj = (dbuf)->db.db_object; \
461 if (__db_obj == DMU_META_DNODE_OBJECT) \
462 (void) strlcpy(__db_buf, "mdn", sizeof (__db_buf)); \
464 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
465 (u_longlong_t)__db_obj); \
466 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
467 "obj=%s lvl=%u blkid=%lld " fmt, \
468 __db_buf, (dbuf)->db_level, \
469 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
473 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
474 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
475 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
476 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
477 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
478 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
482 #define DBUF_VERIFY(db) dbuf_verify(db)
486 #define dprintf_dbuf(db, fmt, ...)
487 #define dprintf_dbuf_bp(db, bp, fmt, ...)
488 #define DBUF_VERIFY(db)
497 #endif /* _SYS_DBUF_H */